A Taste of Agent-Based Modelling

Why ABMs interest me

• Computer Science background
• I like to code
• I like to think about social issues

Psychology is complicated

e.g. Communal motivation involves

• Cultural continuity
• Ostracism
• Social embeddedness
• Social capital
• Commitment mechanisms
• Social media
• Identity fusion
• ………

Psychology is complicated

• How do we combine ideas?
• Which effects are dominant?
• Generalizability

Motivation

• Replication crisis
• Social scientists make poor real world predictions (Grossmann et al. 2024)
• Is human behaviour too complicated for our toolkit?

Where ABMs Can Help

• Combining theories
• Flexible Analyses
• High fidelity
• Personal development?

Parts of an Agent-based Model

Environment

• Network
• Discrete grid
• Continuous space

Agents

Time Step

Parts of an Agent-based Model

Time 0

• Generate environment
• Generate agents and place them in the environment

Time 1, 2, 3, …, Stop

• Agents activate
• The environment can change

What to model?

Evolution

• Simple premise
• Constant thread through theories

Implementing a simple idea

flowchart LR
  A["Fitness"] --> B["Prevalence"]

Software

• Python
• Mesa (Ter Hoeven et al. 2025)
• pandas (team 2025; McKinney 2010)
• seaborn (Waskom 2021)
• Altair (VanderPlas et al. 2018; Satyanarayan et al. 2017)

How does fitness relate to prevalence?

• Each agent has a fitness value
• This fitness value has to affect the agent’s reproduction

How does fitness relate to prevalence?

class Agent:
  self.fitness

  reproduce():
    self.fitness -> probability of success

  step():
    reproduce()

Is simplicity the goal?

My first implementation

class Agent:
  self.fitness
  self.energy

  feed():
    self.fitness -> probability of +2 energy

  reproduce():
    always successful
  
  metabolism():
    self.energy - 1
  
  step():
    metabolism()
    reproduce()
    if self.energy < 0 then remove agent
    else feed()